To get any meaningful information out of unstructured text, you need to establish some context - the more context the better the results.
Not all API calls are equal. The language identifier API call requires very little text to return a language identification while extracting entities on the other hand requires more context (i.e. news articles) to be successful.
The current maximum text you can process with one API call is 50k characters.